(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: To develop workflows that capture and preserve costly information in radiomics and to establish good methodological practices regarding data management. Radiomic activities require costly information such as regions of interest drawn by experts (e.g. radiation oncologists) and textural, morphological and statistical signatures obtained from intensive numerical calculations. Workflows that preserve this costly and valuable information are highly desirable.
Methods: We have designed and implemented workflows aligned with the FAIR principles to manage radiomic data. These workflows use recognized standards, permanent identifiers, and terminologies used by the medical imaging community. More specifically, our workflows rely on the DICOM standard to store not only image-related information but also the context of their capture/generation (who, when, how). All the data are pooled in an open-source DICOM server that is queried by radiomic pipelines.
Result: FAIR-inspired workflows were implemented with two open-source software: 3DSlicer to create DICOM objects and the Orthanc DICOM server to manage these objects. The Quantitative Reporting extension of 3DSlicer was used to generate DICOM segmentation objects as well as the creation context (e.g. operator name). The SlicerRT extension was used to transform segmentations into RTStruct DICOM objects, using the AAPM TG-263 naming convention. Clinical information pertaining to images was also embedded as DICOM Structured Reports and linked to images through unique identifiers. DICOM objects were stored in an Orthanc instance. Radiomic pipelines then interact with Orthanc through its REST API to access images and related information. This data management strategy allows us to efficiently store, understand, and reference data and metadata, and to ensure reproducible research activities.
Conclusion: The developed workflows, relying on the DICOM standard, ensure the quality and reproducibility of radiomics research. Good practices described here allow the creation of well documented, interoperable data sets that can be machine-harvested for large scale radiomics endeavors.
Keywords
Not Applicable / None Entered.
Taxonomy
IM/TH- Informatics: Informatics in Imaging (general)
Contact Email
